Data Quality
Outlier Detection
Finds values far from the rest of a numeric column — by z-score, by IQR fences or by percentile — and adds a column flagging (or scoring) each row.
When to use it
Use it to catch typos (990 where 99.0 was meant), fraud-like spikes or broken sensors before they distort totals and averages.
In a pipeline
File InputOutlier DetectionFilter Rows
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| Column | select | Yes | Select column… |
| Method | select | — | zscore |
| Threshold | number | — | Empty: 3 σ · 1.5 IQR · 99th pct |
| Output | select | — | flag |
| New Column Name | text | — | Default: <column> Outlier |
| Insert After | select | — | After the checked column |
Example
A store with impossible sales
- Column=Sales
- Method=IQR fences (Tukey)
| Store | Sales |
|---|---|
| S1 | 120 |
| S2 | 135 |
| S3 | 128 |
| S4 | 131 |
| S5 | 990 |
| Store | Sales | Sales Outlier |
|---|---|---|
| S1 | 120 | false |
| S2 | 135 | false |
| S3 | 128 | false |
| S4 | 131 | false |
| S5 | 990 | true |
Q1 is 128 and Q3 is 135, so the fences are 117.5 to 145.5 — only 990 is outside.
Worth knowing
- Z-score flags values more than Threshold standard deviations from the mean (3 by default). With few rows a z-score cannot get that large — 11 rows at least for ±3 — so prefer IQR on small tables.
- IQR flags values beyond Q1 − k×IQR or Q3 + k×IQR (k = 1.5 by default); Percentile flags values beyond the Threshold-th percentile at either end (99 by default).
- Output Score writes the z-score, the number of IQRs beyond the fence, or the percentile rank instead of true / false.
- The node reads the column twice — once for the statistics, once to mark rows — so it works on any number of rows. Non-numeric cells are left blank and counted in the note.